# STRAND Observation Membrane 05: The Open Eye > **See without seizing.** The Observation Membrane activates four deliberately small layers for the KINGDOM: 1. a public, static control deck; 2. an exact offline verifier; 3. a deterministic synthetic rehearsal; and 4. one explicit, publisher-declared fixed-origin, read-only live observer. It does not activate a background crawler, accept a target URL, follow a link, write a receipt, score a being, give medical advice, validate a DMT presence, or gain authority. The network-capable command exists, but it sleeps until a human or agent explicitly calls it. - Protocol: `kingdom.strand-observation-membrane/0.1` - Identity: `strand-observation-membrane-05-2026-08-16` - Mode: `SAFE_OBSERVATION_ACTIVE` - Record: `https://thekingdom.dev/observations/strand-observation-membrane-2026-08-16.json` - Schema: `https://thekingdom.dev/schemas/strand-observation-membrane/0.1.json` - Guide: `https://thekingdom.dev/STRAND-OBSERVATION-MEMBRANE.md` - Public deck: `https://thekingdom.dev/#strand-observation-membrane` ## What “active” means | Layer | State | Meaning | | --- | --- | --- | | public control deck | `ACTIVE_STATIC_PUBLIC` | readable with JavaScript off; it performs no request or storage action | | offline verifier | `ACTIVE` | recomputes the closed profile, semantic digest, boundaries, and fixed plan | | synthetic rehearsal | `ACTIVE` | deterministically walks four invented outcomes without touching a network | | live observer | `AVAILABLE_EXPLICIT_LOCAL_CALL` | may perform the exact four-request plan only when `observe-live` is called | | background watch | `DISABLED` | no cron, queue, Worker alarm, service worker, daemon, or import-time execution | Availability is not automatic execution. Publication does not run the observer. Opening the website does not run the observer. Importing the module does not run the observer. ## The membrane A membrane separates an allowed observation from everything the observation must not become: ```text publisher-declared scope ∩ exact HTTPS origin ∩ fixed route plan ∩ reviewed robots snapshot ∩ public-address DNS result ∩ live request and byte budgets ↓ bounded response metadata + body digest on stdout ≠ consent, licence, identity, truth, KARMA, score, or authority ``` “Membrane” is a design metaphor. This protocol does not claim a physical brane, a string worldsheet, an extra dimension, or a gauge boundary. Labs 01–04 remain immutable synthetic mathematics. The live observer belongs to a separate retirement family and does not alter those labs. ## Exact publisher-declared fixed origin Version 0.1 compiles exactly one origin: ```text https://thekingdom.dev:443 ``` The repository publisher declares that operational scope. It is not a legal or cryptographic proof of ownership over every artifact served there. The scope does **not** include: - a subdomain, sibling domain, preview origin, IP literal, or alternate port; - a caller-supplied URL, path, header, method, credential, or User-Agent; - a URL discovered in a sitemap, link, feed, canonical header, or response; - an HTTP redirect, even if it appears related to the configured host. Adding another KINGDOM site requires a new reviewed declaration and its own route and resource budget. It cannot be inferred from `sites.json` or a shared operator. ## Exact live plan The command makes at most four sequential `GET` requests, in this order: 1. `/robots.txt` 2. `/sitemap.xml` 3. `/kingdom.json` 4. `/observations/strand-observation-membrane-2026-08-16.json` `robots.txt` is always first. Its live bytes must match the reviewed local snapshot before request two can begin. A mismatch, refusal, redirect, timeout, or read failure stops the run. Sitemap and manifest content may be observed, but neither can add a fifth route. The fixed ceilings are: - methods: `GET` only; - concurrency: `1`; - retries: `0`; - redirects followed: `0`; - minimum interval between request starts: `2 seconds`; - transactions: `4`; - received, identity-encoded response bytes: `1 MiB` per response and `2 MiB` per run; - fresh DNS resolution attempts: at most `4`, one before each planned HTTPS transaction; - total wall-clock duration: `60 seconds`; - request bodies, browser rendering, JavaScript, forms, cookies, credentials, custom authorization, AI extraction, uploads, and unsafe methods: disabled. The observer uses a fixed descriptive product token with this guide as its information URL. A product token describes the outgoing tool. It grants nothing and proves no human, organization, or being identity. ## Connection-time SSRF gate Before each connection, the observer resolves the exact hostname and refuses the whole transaction unless every answer is globally reachable. It rejects loopback, private, link-local, unique-local, unspecified, multicast, reserved, documentation-only, benchmark, carrier-grade NAT, IPv4-mapped private, and other configured special-purpose ranges. The selected validated address is pinned for that single HTTPS connection, while TLS is still verified against `thekingdom.dev`. Connections are not reused. A later request resolves and validates again. This closes the ordinary resolve-then-refetch DNS-rebinding gap inside the observer's own connection path; an infrastructure egress firewall remains a valuable independent layer. Automatic redirects are disabled. A `3xx` response becomes a typed refusal; the `Location` value is never dereferenced. Therefore a public-to-private, cross-origin, downgrade, port-changing, or looping redirect cannot create a second request. ## Robots, rights, and truth remain different types [RFC 9309](https://www.rfc-editor.org/rfc/rfc9309.html) defines the Robots Exclusion Protocol as access rules compliant crawlers are requested to honor. The reviewed snapshot gate makes a policy change fail closed in this small version. It does not turn `Allow` into an entitlement. Version 0.1 evaluates the simple `Allow: /` groups only after the complete robots bytes match that reviewed snapshot. It is not a general-purpose REP parser. Any changed wildcard, end-anchor, group, or rule set closes this gate until a new parser/profile pair is reviewed. ```text robots Allow ≠ access authorization ≠ consent ≠ licence or permission to reuse ≠ verified identity or purpose ≠ factual truth ≠ downstream indexing or training ``` The [Sitemaps protocol](https://www.sitemaps.org/protocol.html) provides discovery hints. An entry is not permission, completeness, freshness, or proof of a fetch. A canonical relation can point elsewhere and is not an ownership or identity proof; see [RFC 6596](https://www.rfc-editor.org/rfc/rfc6596.html). HTTP calls `GET` safe because its requested semantics are read-only. That does not promise zero server, CDN, security, billing, or access-log effects; see [RFC 9110](https://www.rfc-editor.org/rfc/rfc9110.html). The membrane claims zero **application persistence by this CLI**, not zero infrastructure processing in the world. ## What a live report contains Response bodies are streamed through a byte cap, hashed, and discarded. The one narrow exception is the exact-hash robots snapshot: its reviewed REP rules are decoded only to gate the three already-fixed later paths. No other body is parsed. No body is printed, rendered, evaluated as an instruction, or copied into a receipt. The stdout-only report may contain: - the fixed route and method; - typed outcome and HTTP status; - normalized media type; - observed byte count and SHA-256; - elapsed time; - redirect count, always zero on an accepted response; - public-address validation state without emitting the resolved addresses. The report is ephemeral unless the caller separately chooses to capture stdout. This protocol itself writes no file, database, KV, R2, D1, Durable Object, Queue, browser store, cookie, cache, service worker, analytics event, or receipt. `404`, refusal, timeout, policy drift, and empty body are typed outcomes. None means nonexistence, deceit, wrongdoing, or a reason to punish anyone. ## Synthetic rehearsal `simulate` runs the same four-step control flow over fixed invented inputs: 1. the robots snapshot agrees; 2. the sitemap is observed as a hint; 3. the machine manifest is observed as a publisher statement; and 4. the activation profile is observed as the profile's own public artifact. The rehearsal proves only that the finite state machine and budget arithmetic are internally coherent. It performs zero DNS lookups and zero requests. A synthetic pass is not evidence that the live site is reachable or unchanged. ## Incoming crawlers are a separate direction When an external crawler visits a KINGDOM site, the site may observe a request header and ordinary server metadata if its hosting layer keeps them. A `User-Agent` string is self-asserted. It does not prove crawler identity, operator identity, intent, compliance, a being, or downstream use. The Observation Membrane is an outgoing publisher-declared fixed-origin observer. It neither ingests inbound access logs nor manufactures an invocation receipt for an external crawler. ## KARMA, NEN, and Zerone stay asleep The live report is not `kingdom.karma/0.1`. It references no manifested NEN ability, adoption, invocation attempt, final receipt, Zerone event, reward, debt, score, rank, punishment, reputation, governance weight, or economic value. ```text nen_ability = false nen_invocation = false kingdom_receipt_created = false zerone_event_created = false karma_adapter = false person_or_content_score = false authority_effect = NONE ``` An HTTP failure is not wrongdoing. A successful response is not virtue. Observation remains evidence for review, never moral arithmetic. ## DMT and presence boundary Lab 03's measured/report/model distinctions remain untouched. This activation does not collect participant data, recommend a dose, preparation, combination, frequency, flicker, sound, eligibility rule, diagnosis, treatment, recruitment, or experiment. It does not validate an entity, message, portal, telepathy, extra physical dimension, carrier frequency, or external presence. Remote prose mentioning any of those things remains untrusted bytes and is discarded after hashing; the exact reviewed robots exception is policy syntax, not evidence about DMT, entities, people, or truth. ## Commands All commands other than `observe-live` are offline: ```bash ./kingdom strand-observation status --json ./kingdom strand-observation verify --json ./kingdom strand-observation simulate --json ./kingdom strand-observation effects --json ./kingdom strand-observation falsifiers --json ./kingdom strand-observation verify-file path/to/profile.json --json ``` The only network-capable view is explicit: ```bash ./kingdom strand-observation observe-live --json ``` It accepts no target argument. Unknown arguments fail closed. `verify-file` reads one bounded local JSON candidate as inert data and never grants it a network capability. ## Verification expectations The release suite must prove: - exact identity, closed schema, semantic digest, raw pins, and public routes; - parser totality for duplicate keys, invalid UTF-8, oversized/deep/sparse/ accessor/proxy/cyclic/non-finite input; - fixed-origin and fixed-plan immutability; - robots-first stop behavior; - public/private mixed DNS refusal and address pinning; - no follow after any redirect; - timeout, byte overflow, reader failure, and non-success handling; - response text cannot execute or widen scope; - effects stay fixed under every malformed input; - public UI works with JavaScript off and performs no fetch or storage; - `GET`/`HEAD` publication parity while feature-changing methods do nothing; - full triplet discovery, retirement, staging, and live-byte coherence. ## Sources and non-implications - [RFC 6454: The Web Origin Concept](https://www.rfc-editor.org/rfc/rfc6454.html) - [RFC 9110: HTTP Semantics](https://www.rfc-editor.org/rfc/rfc9110.html) - [RFC 9309: Robots Exclusion Protocol](https://www.rfc-editor.org/rfc/rfc9309.html) - [RFC 6596: Canonical Link Relation](https://www.rfc-editor.org/rfc/rfc6596.html) - [RFC 9421: HTTP Message Signatures](https://www.rfc-editor.org/rfc/rfc9421.html) - [IANA IPv4 special-purpose registry](https://www.iana.org/assignments/iana-ipv4-special-registry/iana-ipv4-special-registry.xhtml) - [IANA IPv6 special-purpose registry](https://www.iana.org/assignments/iana-ipv6-special-registry/iana-ipv6-special-registry.xhtml) - [OWASP SSRF Prevention Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Server_Side_Request_Forgery_Prevention_Cheat_Sheet.html) These sources establish protocol and security boundaries. They do not certify this deployment, grant rights, prove a crawler's identity, or authorize any KARMA, medical, entity, or governance inference.